Rotational Symmetry of the Structured Chip/LDB-SSDP Core Module of the Wnt Enhanceosome

Rotational symmetry of the structured Chip/LDB-SSDP core module of the Wnt enhanceosome

Miha Renkoa,1, Marc Fiedlera,1, Trevor J. Rutherforda, Jonas V. Schaeferb, Andreas Plückthunb, and Mariann Bienza,2

aMedical Research Council Laboratory of Molecular Biology, CB2 0QH Cambridge, United Kingdom; and bDepartment of Biochemistry, University of Zurich, 8057 Zurich, Switzerland

Edited by Roeland Nusse, Stanford University School of Medicine, Stanford, CA, and approved September 9, 2019 (received for review July 23, 2019) The Chip/LIM-domain binding protein (LDB)–single-stranded DNA- mediating enhancer function. Consistent with this, chicken Ssdp1 binding protein (SSDP) (ChiLS) complex controls numerous cell-fate and Ssdp2 are required to confer transcriptionally active chro- decisions in animal cells, by mediating transcription of develop- matin on loci controlled by Ldb1 (15), and murine Ssbp2 is re- mental control genes via remote enhancers. ChiLS is recruited to quired for hematopoietic stem cell maintenance, similarly to Lbd1 these enhancers by lineage-specific LIM-domain proteins that bind (18). The molecular basis underlying this intimate cooperation to its Chip/LDB subunit. ChiLS recently emerged as the core mod- between Chip/LDB and SSDP is not known. ule of the Wnt enhanceosome, a multiprotein complex that primes We recently discovered that Chip/LDB and SSDP form a developmental control genes for timely Wnt responses. ChiLS stable complex called ChiLS (Chip/LDB-SSDP), which binds to binds to NPFxD motifs within Pygopus (Pygo) and the Osa/ARID1A NPFxD motifs within the nuclear Wnt signaling factor Pygopus subunit of the BAF chromatin remodeling complex, which could (Pygo) and Osa/ARID1A (17), a chromatin-binding subunit of synergize with LIM proteins in tethering ChiLS to enhancers. Chip/ the BAF transcriptional coactivator complex (19). The Wnt LDB and SSDP both contain N-terminal dimerization domains that enhanceosome is a ChiLS-containing multiprotein complex that constitute the bulk of their structured cores. Here, we report the is tethered to Wnt-responsive transcriptional enhancers via T cell crystal structures of these dimerization domains, in part aided by factors/lymphoid-enhancer binding factors (TCF/LEF) and their DARPin chaperones. We conducted systematic surface scanning by associated Groucho/transducin-like enhancer (TLE) corepres- structure-designed mutations, followed by in vitro and in vivo binding assays, to determine conserved surface residues required sors, to prime linked developmental control genes for timely Wnt BIOCHEMISTRY for binding between Chip/LDB, SSDP, and Pygo-NPFxD. Based on responses (17, 20). The Wnt response of this complex is con- this, and on the 4:2 (SSDP-Chip/LDB) stoichiometry of ChiLS, we ferred by Pygo, which facilitates loading of the Wnt effector β derive a highly constrained structural model for this complex, -catenin via an adaptor called BCL9, thereby promoting tran-

which adopts a rotationally symmetrical SSDP2-LDB2-SSDP2 archi- scriptional activation (ON state) (17, 20), while its rerepression tecture. Integrity of ChiLS is essential for Pygo binding, and our (OFF state) appears to depend on Osa (21). The Wnt enhanceosome mutational analysis places the NPFxD pockets on either side of the model envisages a pivotal role of ChiLS in the assembly and Chip/LDB dimer, each flanked by an SSDP dimer. The symmetry function of the complex, implicating ChiLS as its switch module: Its and multivalency of ChiLS underpin its function as an enhancer chromatin tethering involves a combination of enhancer-associated module integrating Wnt signals with lineage-specific factors to operate context-dependent transcriptional switches that are piv- Significance otal for normal development and cancer. A structural model is presented for the core Chip/LIM-domain Wnt enhanceosome | Chip/LDB1-SSDP | Pygo binding protein (LDB)–single-stranded DNA-binding protein (SSDP) (ChiLS) complex, which mediates long-range enhancer– ertebrate LIM-domain binding protein 1 (LDB1, also known promoter interactions to control the transcription of master Vas NLI or CLIM) and its Drosophila ortholog Chip have regulatory genes with key functions in embryonic develop- pleiotropic functions in controlling embryonic and larval devel- ment, stem cell maintenance, and differentiation along adult opment at multiple stages (1–4), maintenance of intestinal and cell lineages, including erythroid maturation. Because of its hematopoietic stem cells (5, 6), and differentiation along ery- rotational symmetry and multivalency, ChiLS can bind to mul- throid cell lineages (7, 8). Chip/LDB proteins are recruited to tiple combinations of lineage-specific DNA-binding proteins, as remote transcriptional enhancers of pivotal developmental con- well as cofactors responding to extracellular signals such as trol genes by their C-terminal LIM-interacting domain (LID) Wnt, and is thus uniquely poised to integrate these inputs and that binds directly to LIM-homeodomain DNA-binding proteins, translate them into transcriptional ON and OFF states of or to GATA and basic-loop-helix (bHLH) DNA-binding pro- target genes. teins via LIM-only adaptors. They facilitate communication between remote enhancers and proximal promoters (2, 9), likely via Author contributions: M.R., M.F., and M.B. designed research; M.R., M.F., and T.J.R. per- – formed research; J.V.S. and A.P. contributed new reagents/analytic tools; M.R., M.F., T.J.R., looping out intervening sequences (8). These long-range enhancer and M.B. analyzed data; and M.B. wrote the paper. promoter interactions critically depend on, and are mediated by, The authors declare no competing interest. self-interaction of LDB proteins (10, 11) through their N-terminal This article is a PNAS Direct Submission. dimerization domain (DD) (12, 13). This open access article is distributed under Creative Commons Attribution License 4.0 Chip/LDB binds to a single-stranded DNA binding protein (CC BY). (SSDP, also known as single-stranded binding protein, SSBP) Data deposition: The atomic coordinates and structure factors have been deposited in the through the LDB1/Chip conserved domain (LCCD), a short re- Protein Data Bank, www.pdb.org (PDB ID codes 6S9R for SSDP, 6D9T for DD-DARPin3, gion downstream of the DD (14, 15). The loss-of-function phe- and 6S9S for DD-DARPin10). notypes of Drosophila chip mutants closely resemble those of ssdp 1M.R. and M.F. contributed equally to this work. mutants in various developmental contexts, although the former 2To whom correspondence may be addressed. Email: [email protected]. tend to be stronger and more pleiotropic than the latter (14, 16), This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. especially in the early embryo (17). These phenotypic similarities 1073/pnas.1912705116/-/DCSupplemental. indicate an intimate cooperation between Chip and SSDP in First published September 30, 2019.

www.pnas.org/cgi/doi/10.1073/pnas.1912705116 PNAS | October 15, 2019 | vol. 116 | no. 42 | 20977–20983 Downloaded by guest on September 24, 2021 proteins (including LIM proteins, Pygo, TCF/LEF-associated 1 of which yielded diffracting crystals that enabled us to determine Groucho/TLE ,and the BAF complex) that determine jointly the structure of SSDP-N (SI Appendix,TableS1). the ON and OFF states of Wnt-controlled downstream genes (17, 20). Importantly, the stoichiometry of the minimal stable Structure of the SSDP-N Dimer. The asymmetric crystallographic α α ChiLS complex is 2:4 (Chip/LDB:SSDP), as determined by size- unit contains a classic LisH dimer-fold in which 2 -helices ( 1, α B exclusion chromatography followed by multiangle light-scattering 2) (Fig. 1 ) associate to form an antiparallel 4-helix bundle (SEC-MALS) of purified recombinant proteins (17), implying that (22), the hydrophobic interface of which is strengthened by additional hydrogen bonds in its periphery (Fig. 1C). A third helix ChiLS contains 1 Chip/LDB dimer and 2 SSDP dimers. α α ′ Here, we report the crystal structures of the DD of SSDP and ( 3) interacts with its counterpart ( 3 ) from the opposite dimer XLdb1, in the latter case with the help of DARPin (designed to form an X-shaped structure, tucked under the 4-helix bundle. ankyrin repeat proteins) chaperones that mask a hydrophobic It thus resembles TBL1, a subunit of a transcriptional co- repressor complex and the only other LisH-containing protein surface patch of the DD required for its SSDP binding. Sys- known to form tetramers (23), except that the lengths and angles tematic structure-led mutagenesis of conserved solvent-exposed between α3 and α3′ differ in the 2 structures (SI Appendix, Fig. residues of Chip/LDB and SSDP followed by in vitro and in vivo S1). As in other LisH dimers, the dimer interface is extensive binding assays enabled us to generate an interaction map, and to 2 (burying ∼2,500 Å ), which explains why SSDP forms obligate derive a highly constrained structural model for ChiLS. Its dimers, and why SSDP monomers are not detectable (17) (see rotationally symmetric SSDP2-LDB2-SSDP2 architecture with 2 also below). Indeed, the Kd for LisH-mediated self-association of predicted NPFxD-binding pockets on either side underscores its Lis1 was reported to be subfemtomolar (22). function as an integrating core module of the Wnt enhance- Within the crystal lattice, symmetry-related dimers associate osome and other ChiLS-containing enhancer-binding complexes. via 2 perpendicular pairs of antiparallel α3 helices. The resulting interlocked bundle of 4 α-helices bury ∼1,400 Å2 (Fig. 2A), and is Results held together entirely via hydrophobic contacts, with valine 65 SSDP contains an N-terminal LisH domain known to form ob- (V65) occupying a central position in the tetramer interface (“α3 ligate dimers in other proteins (22), followed by a highly con- interface,” below). The same tetramer was observed for human served extension and a long nonconserved C-terminal tail likely SSBP2 (whose structured region differs from fly SSDP-N only by to be disordered (Fig. 1A). The extended N-terminal dimeriza- 2 semiconserved residues) (Fig. 1B) (24) whose structure was tion domain (called SSDP-N below; also known as the LUFS reported during the final stages of our work. Intriguingly, some domain) (14) can form stable tetramers in vitro following bac- of the hydrophobic residues in the loop between α2 and α3 also terial expression (17). To determine the structure of SSDP-N, we mediate association between symmetry-related dimers in the optimized its boundaries and succeeded in obtaining crystals for crystal, resulting in alternative tetramer configurations (Fig. 2B). SSDP1–86 that diffracted to 2.4 Å. As we were unable to use Although these loop-mediated interactions bury smaller inter- molecular replacement with the LisH domain as a search model, faces (602 Å2 and 624 Å2, respectively; ”loop interface,” below), we generated 5 different selenomethionine (SeMet)-labeled mutants, they proved to be functionally relevant in vivo (see below). We

Fig. 1. Structure of the SSDP dimer. (A) Cartoons of SSDP and Chip/LDB and their domains (numbers, Drosophila SSDP and Chip). (B) Sequence alignment of the N-terminal dimerization domain of Drosophila (Dm), human (Hs), and Trichoplax adhaerens (Ta) SSDP, with secondary elements indicated above; bold indicates amino acid variations between Hs SSBP2 and Dm SSDP (structured region only); colored indicates crucial mutated residues. (C) Ribbon representations of SSDP dimer, with helices labeled; (Left Inset) dimerization interface between α2 and α2′, with key residues shown in stick, and hydrogen bonds (between α2′ E43 and main chain α2 atoms) as yellow dashed lines; (Right Inset) α3/α3′ dimerization interface, with key hydrophobic residues shown in stick.

20978 | www.pnas.org/cgi/doi/10.1073/pnas.1912705116 Renko et al. Downloaded by guest on September 24, 2021 2 hydrophobic residues in the loop interface (I50E L52E, called ML) that mediate mutual interactions in the crystals (Fig. 1B). We also included a previously designed triple mutation of hydrophobic residues flanking V65 (F66E L69A Y70A, called Mα3) that blocks SSBP2-N tetramerization (24). These mutations were introduced into Lip-SSDP1–92 and their self-interaction was tested by SEC-MALS. As expected (24), Mα3 blocks tetramerization of SSDP-N, but V65K does not (Fig. 2C), likely because this single mutation is too weak to do so, given the extensive α3 interface. ML also reduces tetramerization of SSDP-N significantly (Fig. 2C). We conclude that both modes of SSDP-N tetramerization can occur, with α3 being the preferred interface. Tetramerization likely occurs during bacterial expression since SSDP-N dimers and tetramers are stable in solution for >8 h, and do not interconvert following purification. Next, we designed another 16 substitutions in conserved solvent- exposed residues that should not perturb the fold of SSDP-N (SI Appendix,Fig.S2A), and introduced these (plus V65K) into full- length V5-tagged SSDP. We then coexpressed these with WT FLAG-tagged SSDP in HEK293T cells from which we deleted LDB1 and LDB2 by CRISPR engineering (below, double-knockout [DKO] cells) (SI Appendix, Fig. S3), to test their mutual association by coimmunoprecipitation (co-IP) in the absence of endogenous LDB. As expected, SSDP-FLAG co-IPs efficiently with WT and most mutant SSDP-V5, except for I50E, L52E, P55E, and F58E (each altering the loop surface) (Fig. 2B and SI Appendix, Fig. S2A), which do not coimmunoprecipitate at all.

Mα3 produces a reduced co-IP signal, partly because this triple- BIOCHEMISTRY mutant is somewhat unstable in cells (Fig. 2D and SI Appendix, Fig. S2B). We conclude that full-length SSDP dimers also self- associate in cells upon overexpression, but do so entirely through their loop interface. This implies that the SSDP C terminus blocks self-association of full-length SSDP via its α3 surface.

Structure of the LDB Dimer. Next, we purified fragments spanning the DD or DD-LCCD from human LDB1, linked to a cleavable His6 and Lipoyl (Lip) tag, but could not crystallize these. Screening LDB orthologs from different species, we found that Xenopus laevis Ldb1 (XLdb1) (1) yielded the most stable protein after removal of solubility tags. XLdb1 is closely related to human LDB1 (SI Appendix, Fig. S4), but we did not succeed in obtaining diffracting crystals with this protein either, largely because the DD and DD-LCCD undergo nonspecific aggregation over time. We therefore resorted to chaperone-based ap- proaches, selecting nanobodies (25) or high-affinity DARPins, genetically engineered antibody-mimetics based on consensus ankyrin repeat proteins and selected from diverse synthetic libraries (26), against the purified DD or ChiLS complex (SI Appendix, Supplementary Methods). None of 35 selected nanobodies facili- tated crystallization, but we obtained DD crystals with 5 of 15 selected DARPins (SI Appendix,Fig.S5) that diffracted to 2.0 to 2.59 Å (SI Appendix, Table S1), depending on the DARPin (SI Appen- Fig. 2. Two modes of SSDP tetramerization. (A and B) Tetramerization of dix A SSDP through (A) α3 interface or (B) loop interface (between α2 and α3), , Fig. S6 ). None of these 5 DARPins bind to the DD-LCCD with key mutated residues in stick (same color in all panels). (C) SEC-MALS nor to the assembled ChiLS complex.

profiles of WT and selected Lip-SSDP1–92 mutants, as indicated; numbers in Next, we determined the crystal structures of the DD (XLdb1 Upper,Mr determined by MALS (corresponding to SSDP2 and SSDP4;expec- 20-200) bound to DARPin2, -3, -5, -7, or -10 (SI Appendix, Fig. ted Mr, 47 and 94 kDa, respectively). (D) co-IP assays of selected SSDP mutants S6). Remarkably, each of the 5 DARPins (despite their different in transfected LDB1/2 DKO cells (see also SI Appendix, Figs. S2 and S3). sequences) recognizes the same lateral surface patch of the DD, perpendicular to its dimerization interface and facing outwards SI Appendix B α SI Appendix ( , Fig. S6 ). This patch is highly hydrophobic (e.g., note that TBL1 tetramerizes via 2( , Fig. S1) via Y81, I83, and L87 engage in direct DARPin contacts), which ex- residues that are not conserved in SSDP. plains why its masking by DARPins was crucial to prevent nonspecific aggregation of the DD via this patch (called “hydrophobic Alternative Modes of Self-Interaction of SSDP Dimers. To test the patch,” below). functional relevance of the different modes of SSDP self-interactions, The DD adopts a cone-shaped α+β barrel-fold, composed of a we designed repelling amino acid substitutions in solvent-exposed long, highly curved antiparallel β-sheet (formed by β1to6) hydrophobic residues in the 2 alternative interfaces of SSDP-N, whose cavity is filled by 3 α-helices (α1toα3) that run roughly mutating a centrally located valine in the α3 interface (V65K), and parallel to the β-sheet (Fig. 3A). This structure is capped by a

Renko et al. PNAS | October 15, 2019 | vol. 116 | no. 42 | 20979 Downloaded by guest on September 24, 2021 Fig. 3. Structure of the DD. (A and B) Ribbon representations of (A) the DD monomer (with α-helices labeled) and (B) the DD dimer; vertical line is the symmetry axis; blue is the DARPin-binding patch. (C) Superimposition of the DD (wheat) and NTF2 (green; 1JB5) (see also SI Appendix, Fig. S7).

short C-terminal extension at the apex of the DD, formed by 2 block binding between DD-LCCD1 and SSDP. Indeed, 2 of intertwined α-helices (α4andα5) that contribute to the dimer in- them do so (L87D R90D and Y81D L87D R90D) (SI Appendix, terface (Fig. 3B). The whole dimer interface is extensive (burying Table S2). Since Y81, L87, and R90 are direct DARPin-interacting ∼1,500 Å2), which explains why the DD dimer is highly stable in solution. Curiously, the closest structural relative of the DD is scytalone dehydratase (27) (rmsd ∼2 Å), a bacterial ketosteroid isomerase (SI Appendix,Fig.S7A). The DD also resembles the fold of other members of this enzyme family (28), which are among the most efficient enzymes known (29). The DD fold is also found in eu- karyotes; for example, in nuclear transport factor-2 (NTF-2) (30) (rmsd ∼2.2 Å) (Fig. 3C). Whether any of these structural similarities have functional significance is unclear. The oligomeric state of the DD folds is variable, ranging from monomeric to tetrameric, and their dimerization modes also vary, with some folds dimerizing via their curved β-sheets (SI Appendix,Fig.S7B). However, none of the known DD folds contain an intertwined apical extension, as seen in the XLdb1 DD (Fig. 3B and SI Appendix,Fig.S7B). This apex may help to determine the dimerization mode of the DD, and stabilize the Chip/LDB dimer, given that a region spanning this apex is essential for β-globin transcription during erythroid maturation (albeit not for LDB1 dimerization per se, as the DD core may be partially competent to dimerize by itself) (11).

Binding Between Chip/LDB and SSDP. Next, we attempted to crystallize the ChiLS complex, using bicistronic coexpression of SSDP1–92 and DD-LCCD from XLdb1. LCCD is a conserved 49- amino acid stretch (Fig. 1A and SI Appendix, Fig. S4) required for binding to SSDP (14, 15). It spans 2 predicted α-helices (α6 and α7), whereby α6 is separated from the DD by a linker of 9 conserved residues (SI Appendix, Fig. S4). We designed 2 DD fragments with C-terminal extensions, namely DD-LCCD1 (20- 244, spanning α6 and α7) or DD-LCCD2 (20-226, spanning only α6). Both form a stable complex with SSDP1–92, implying that α6 suffices for SSDP binding at high protein concentrations during bacterial expression. We purified both complexes, but did not obtain diffracting crystals despite testing a wide range of conditions, as well as cocrystallization with nanobodies or DARPins. We therefore took an alternative approach, namely systematic mutational analysis of conserved solvent-exposed residues of Fig. 4. Mutual interactions between Chip/LDB, SSDP, and Pygo. (A) Func- each ChiLS dimer subunit (Fig. 4A), to identify their surfaces tionally relevant solvent-exposed residues in the DD (Chip residue numbers in parenthesis), tested for interaction with SSDP and Pygo. (B) Surface rep- required for mutual interaction. Since DARPin binding to their B resentation of DD, with hydrophobic patch mediating interaction with SSDP cognate patch of the DD (Fig. 3 ) blocks its interaction with highlighted; Y81, L87, and R90, DARPin-binding residues; C197 marks the C SSDP, we initially focused on this hydrophobic patch, designing terminus of the structured part of the DD, and start of LCCD. (C)co-IPassaysof a set of repelling mutations (SI Appendix, Fig. S8), which might selected mutants in transfected HEK293T cells (see also SI Appendix,TableS3).

20980 | www.pnas.org/cgi/doi/10.1073/pnas.1912705116 Renko et al. Downloaded by guest on September 24, 2021 residues (Fig. 4B), this implicates the hydrophobic patch in the whether excess unlabeled Lip-NPFDD peptide from Pygo can interaction between SSDP and the DD. compete for binding of purified ChiLS to an 15N-labeled 26 15 Next, we designed additional mutations in the apical, lateral, aminoacidpeptidefromOsa( N-Lip-NPFEDOsa), which was the and basal surfaces of the DD dimer, in LCCD and its linker to case (SI Appendix,Fig.S10). This implies that the same ChiLS DD (SI Appendix, Fig. S8), and we also generated internal de- pocket can accommodate the NPFxDmotifofeitherChiLSligand. letions that remove LCCD α6orα7(Δα6, Δα7), for co-IP assays in transfected HEK293T cells aimed at testing the binding between A Highly Constrained Structural Model of ChiLS. Next, we con- full-length proteins. To minimize cross-reaction with endogenous structed a model of the ChiLS complex, taking into account its LDB, we introduced these mutations into Chip-V5FLAG, and constituent dimer structures (Figs. 1C and 3B), the results from coexpressed them with SSDP-V5. This confirmed the functional our mutational analysis (Fig. 4 and SI Appendix, Tables S2 and importance of Y81, L87, and R90, but also identified additional S3) and the previously determined 4:2 (SSDP:Chip/LDB) stoi- residues near the hydrophobic patch as critical for co-IP with SSDP chiometry of the ChiLS complex (17). The only ChiLS configu- (Fig. 4C and SI Appendix,TableS3), marking a contiguous hy- ration that is consistent with all our data corresponds to an drophobic surface in the DD (Fig. 4B). In contrast, repelling point SSDP2-LDB2-SSDP2 architecture with rotational symmetry (Fig. mutations in the apical or basal surface of the DD did not affect 5A). In this model, the lateral hydrophobic surface patches of co-IP between Chip and SSDP (SI Appendix, Table S3). We con- each DD subunit (Fig. 5A, blue) interface with their downstream- clude that a lateral hydrophobic surface patch of the DD is crucial adjacent LCCD α6 (Fig. 5A, orange) and the α3 surface of an for its association with SSDP. This implies that a single Chip/LDB SSDP dimer (Fig. 5A, dark cyan). Because of its inherent rota- dimer can accommodate 2 SSDP dimers. tional symmetry, the complex can accommodate two NPFxD li- In addition to this hydrophobic patch, we also found that α6is gands (Fig. 5A, gray rods), through opposite pockets whose crucial for co-IP between Chip and SSDP, while Δα7 also re- positions are defined by proximity to the LCCD α6 methionines duces it; indeed, mutation of a single methionine in α6 (M406R) (Fig. 5A, orange sticks) and SSDP F58 (Fig. 5A, red patch). reduces co-IP between the 2 proteins to background levels (Fig. To corroborate our model, we asked whether a minimal 4C). M402E also blocks co-IP, and L405D reduces it, as does a LCCD fragment could bind SSDP, and if so, whether this would double mutation in α7; however, none of the mutations in the be sensitive to mutations in its α3 or loop surface. We therefore conserved linker between the DD and LCCD affect co-IP (SI conducted SEC-MALS of Chip-LCCD (residues 384 to 436, of Appendix,TableS3). Thus, α6 is essential for binding between which 47 are identical in human LDB1) (SI Appendix, Fig. S4)

Chip and SSDP, and M402 and M406 may directly contact SSDP. tagged with maltose binding protein (MBP) coexpressed with BIOCHEMISTRY α Consistent with this, a 10-amino acid stretch spanning these me- WT, V65K, M 3, or ML Lip-SSDP1–92. As a further control, we thionines in Chip is essential for its binding to SSDP (14), as is a also included a triple mutant of LCCD (M402E L405D M406R, similar 6-amino acid stretch in chicken Ldb1, whereby the latter is Mα6) expected to block binding to SSDP (Fig. 4C). Indeed, the also critical for motoneuron specification in the chicken embryo WT proteins form predominantly a single complex eluting in 1 (15), indicating the physiological importance of LCCD α6. Either main SEC peak, containing similar amounts of LCCD and SSDP α6 constitutes the SSDP-binding surface of Chip/LDB, together as judged by PAGE of the corresponding fractions (Fig. 5B). = with the lateral hydrophobic patch, or this patch binds primarily to Basedonitsmolarmass(Mr 190kDa)asdeterminedby α6, which in turn binds to SSDP, perhaps the more likely scenario, MALS, the complex in this peak most likely corresponds to for reasons discussed below. In contrast, α7 appears to be auxiliary LCCD-SSDP4-LCCD: That is, 2 LCCD-SSDP2 complexes in mediating the Chip/LDB-SSDP interaction. interacting via the SSDP loop interface (Fig. 5B, cartoon). We To define the Chip-binding surface of SSDP, we conducted also observed a smaller peak (Mr = 90 kDa), likely corre- co-IP assays between WT Chip-V5FLAG and our panel of 17 sponding to LCCD-SSDP2, in addition to minor peaks with SSDP-V5 mutants (Fig. 2D and SI Appendix, Fig. S2A), following considerably higher molar masses, likely corresponding to un- coexpression in HEK293T cells. Most SSDP mutations have no specific aggregates (Fig. 5B, asterisks). As expected, none of the effect (including ML), but D68R reduces co-IP, as does Mα3, mutants show the 190-kDa complex: SSDP Mα3 neither binds to while V65K essentially blocks co-IP (Fig. 4C). We conclude that LCCD nor tetramerizes, and so only forms dimers; SSDP V65K SSDP binds to Chip/LDB via its α3 surface (Fig. 2A). and LCCD Mα6 cannot bind to their partner subunits, and so the 2 main peaks observed with these mutants correspond to LCCD Dependence of Pygo Binding on ChiLS Complex Assembly. Next, we and SSDP4 (Fig. 5B, cartoons). Importantly, SSDP ML can in- tested our SSDP and Chip mutants for interaction with Drosophila teract with LCCD through α3, and so forms LCCD-SSDP2 Pygo upon coexpression in HEK293T cells. This interaction is (possibly in addition to coeluting SSDP4) (Fig. 5B). This con- relatively weak, and not easily detectable by co-IP (17). However, firms that 1) LCCD can bind to SSDP directly via α6, and 2) we consistently obtained a co-IP signal between Pygo and WT SSDP dimers can undergo mutual interactions via their loop Chip and SSDP, but with none of the mutants that block the Chip– surfaces if bound to LCCD. Thus, in the presence of LCCD, the SSDP interaction (Fig. 4C and SI Appendix, Table S3). This sug- α3 surface of SSDP prefers to bind to LCCD rather than itself. gests that the integrity of the ChiLS complex is essential for Pygo binding, consistent with our previous evidence that peptides Discussion spanning NPFxD from Pygo or human Pygo2 bind to the assem- Our work has led to a highly constrained structural model of ChiLS, bled recombinant ChiLS complex but not to its subunits alone the core complex of the Wnt enhanceosome (17, 20). Its rotational (17). Notably, F58E blocks Pygo binding to ChiLS while it barely symmetry implies that ChiLS contains 2 structurally identical NPFxD- affects binding between Chip and SSDP, which suggests that SSDP binding pockets, each bordered by an SSDP dimer and LCCD α6 F58 contributes critically to the NPFxD-binding pocket of ChiLS. (and possibly DD residues) (Fig. 5A), and each binding either Pygo2 To corroborate our co-IP results with in vitro binding assays or Osa NPFxD (SI Appendix,Fig.S10). Thus, 1 single ChiLS core between recombinant proteins, we used NMR spectroscopy (17), complex, via its 2 NPFxD pockets, could accommodate Pygo as well probing an 15N-labeled 27 amino acid peptide spanning NPFED as the Osa/ARID1 subunit of the BAF complex simultaneously. 15 from human Pygo2 ( N-Lip-NPFEDPygo2) with purified recombinant This is consistent with the notion that Pygo and the BAF complex ChiLS. This revealed that WT ChiLS interacts with 15N-Lip- are constitutive components of the Wnt enhanceosome (20). NPFEDPygo2, even without LCCD α7(SI Appendix,Fig.S9), in- ChiLS is also found in other enhancer-binding complexes, dicating that the latter is dispensable for ChiLS binding to NPFxD most notably those binding to the remote LCR enhancer that at high protein concentration. We also used this assay to determine controls β-globin genes during erythroid maturation (8, 18, 31),

Renko et al. PNAS | October 15, 2019 | vol. 116 | no. 42 | 20981 Downloaded by guest on September 24, 2021 but also to transcriptional enhancers of developmental control genes (1–4), and of genes that control stem cell maintenance (5, 6) and normal as well as malignant erythroid differentiation (7, 32, 33). N-terminal domains of Chip/LDB proteins self-associate (12, 13), and this property is crucial for mediating long-range interactions between remote enhancers and proximal promoters (2, 8, 10, 11). The DD dimer, as revealed by our structural analysis (Fig. 3B), provides the molecular basis for this function of Chip/LDB in mediating long-range enhancer– promoter interactions. From its structure, it is difficult to see how DD could oligomerize, as previously proposed (2, 8, 13), but we note that recombinant DD has a tendency aggregate in vitro, likely via its hydrophobic DARPin-binding patch. Recombinant SSDP-N clearly tetramerizes via its α3 surface (24) (Fig. 2 A and C). However, in our co-IP assays involving expression of full-length proteins in cells, SSDP uses its α3 surface exclusively for Chip/LDB binding, while it self-associates via its loop surface (Fig. 2C). Furthermore, our evidence suggests that LCCD binding to SSDP may promote loop-mediated self- association of SSDP (Fig. 5B), possibly by inducing a confor- mational change of its loop surface. Indeed, Chip/LDB-bound SSDP dimers that self-associate via their loop interfaces (Fig. 2 B–D) could promote dimerization of the core ChiLS complex (Fig. 5B), assembling higher-order oligomers that could be in- strumental for the function of ChiLS in mediating long-range enhancer–promoter contacts. Whatever the case, our results strongly support the notion that SSDP is essential for the function of Chip/LDB proteins in transcriptional activation by remote enhancers (14, 17, 18, 31, 34, 35). Our structural model of ChiLS provides mechanistic insight into how ChiLS-containing enhancer complexes integrate multiple inputs from signaling and lineage factors: Because of its symmetrical SSDP2-Chip/LDB2-SSDP2 architecture, a single ChiLS core complex can bind simultaneously to 2 different NPFxD ligands (17, 20) (Fig. 5A)andto2setsofdistinctenhancer-binding proteins via its LID (binding to LIM-containing proteins, or GATA and bHLHfactors)(3,4,7,8).Therefore,ChiLSisuniquelypoisedas an integrating core module of multiprotein complexes that are tethered to transcriptional enhancers by specific combinations of DNA-binding proteins and their associated signal-responsive cofactors. Furthermore, by exchanging some of these factors, ChiLS can switch enhancer complexes between ON and OFF states. Materials and Methods Plasmids. Plasmids used for cell-based assays and bacterial expression were

used as described previously (17), including His8 (for DARPins), His6-Lip (for XLdb1 and SSDP mutants), and bicistronic expression vectors (for coex-

pression of SSDP1–92 and MBP-LCCD).

Protein Purification. Proteins were expressed in BL21(DE3)-RIL, and purified with Ni-NTA resin followed by gel filtration. For crystallization, the solubility tags were removed by gel filtration following digestion with tobacco etch virus protease.

Functional Assays in Human Cells. HEK293T cells were grown in DMEM, sup- plemented with 10% FBS and transfected in 6-well plates with PEI. One- hundred nanograms per well of Chip, 400 ng per well of SSDP, and 500 ng per well of Pygo constructs were used for all transfections. Fig. 5. Structural model of the ChiLS complex. (A) Structural model of ChiLS, derived from stoichiometry of complex and mutational analysis (see text); DARPin Selection. Ribosome display selections against the DD or ChiLS complex (Upper) side view; (Lower) view from top, revealing rotational symmetry of were carried out as detailed in SI Appendix, Supplementary Methods. ChiLS and its binding sites for SSDP and NPFxD (whose precise positions and angles relative to the DD are arbitrary in this model). SSDP dimers (cyan) NMR Spectroscopy. Proteins were expressed in minimal medium supple- were docked manually onto the DARPin-binding patch (blue) in each lateral mented by 15N-ammonium chloride and purified as described above. NMR surface of the DD (wheat); (orange) predicted LCCD α6, with SSDP-binding residues (M402, M406) in stick; positions of the pockets for NPFxD (gray rods) are restrained by proximity to LCCD α6 (orange) and SSDP F58 (red);

note that SSDP′ F58 (gray) differs from SSDP F58 (red) regarding its struc- Lip-SSDP1–92; numbers are Mr values determined by MALS (within panels), or tural environment. (B) SEC-MALS profiles of complexes (cartoons above as expected (above panels); asterisks are large aggregates; below, PAGE

panels) formed between coexpressed WT and mutant MBP-LCCDChip and revealing proteins in corresponding preparative SEC fractions.

20982 | www.pnas.org/cgi/doi/10.1073/pnas.1912705116 Renko et al. Downloaded by guest on September 24, 2021 spectra were recorded with a Bruker Avance III spectrometer at 600 MHz 1H, (SI Appendix, Table S1) after growing for several days at 19 °C by the vapor- as described previously (17). diffusion method, and were directly flash-frozen in liquid nitrogen. For determination of structures, see SI Appendix, Supplementary Methods. SEC-MALS. Purified proteins were analyzed with an Agilent 1200 Series chromatography system connected to a Dawn Heleos II 18-angle light-scattering ACKNOWLEDGMENTS. We thank Roger Williams, Chris Johnson, and Melissa detector combined with an Optilab rEX differential refractometer (Wyatt). Gammons for technical advice and discussions; Thomas Reinberg and other Samples were loaded onto a Superdex-200 10/300 gel-filtration column (GE members of the High-Throughput Binder Selection Facility for their contri- Healthcare) at 2 mg/mL and run at 0.5 mL/min in buffer (PBS, 1 mM DTT). Data butions to generating the DARPins used in this study; Jan Steyaert, Els were processed with Astra V software. Pardon, and Thomasz Uchanski for nanobody production (funded by Instruct-ERIC); and the Diamond Light Source (beamlines I04 and I24) and European Synchrotron Radiation Facility (beamline ID23-1) for beamtime and Crystallization. Concentrated proteins (10 to 20 mg/mL) were used for initial assistance. This work was supported by Cancer Research UK (Grants C7379/ screens with ∼1,500 different crystallization conditions in 100 + 100-nL drops A15291 and C7379/A24639, to M.B.) and the Medical Research Council (Grant in a 96-well sitting-drop format. Crystals emerged under multiple conditions U105192713, to M.B.).

1. A. D. Agulnick et al., Interactions of the LIM-domain-binding factor Ldb1 with LIM 19. C. Kadoch, G. R. Crabtree, Mammalian SWI/SNF chromatin remodeling complexes and homeodomain proteins. Nature 384, 270–272 (1996). cancer: Mechanistic insights gained from human genomics. Sci. Adv. 1, e1500447 2. D. Dorsett, Distant liaisons: Long-range enhancer-promoter interactions in Drosophila. (2015). Curr. Opin. Genet. Dev. 9, 505–514 (1999). 20. L. M. van Tienen, J. Mieszczanek, M. Fiedler, T. J. Rutherford, M. Bienz, Constitutive 3. J. M. Matthews, J. E. Visvader, LIM-domain-binding protein 1: A multifunctional co- scaffolding of multiple Wnt enhanceosome components by Legless/BCL9. eLife 6, factor that interacts with diverse proteins. EMBO Rep. 4, 1132–1137 (2003). e20882 (2017). 4. R. Bronstein, D. Segal, Modularity of CHIP/LDB transcription complexes regulates cell 21. R. T. Collins, J. E. Treisman, Osa-containing Brahma chromatin remodeling complexes differentiation. Fly (Austin) 5, 200–205 (2011). are required for the repression of wingless target genes. Genes Dev. 14, 3140–3152 5. I. Dey-Guha, M. Mukhopadhyay, M. Phillips, H. Westphal, Role of ldb1 in adult in- (2000). testinal homeostasis. Int. J. Biol. Sci. 5, 686–694 (2009). 22. M. H. Kim et al., The structure of the N-terminal domain of the product of the – 6. L. Li et al., Nuclear adaptor Ldb1 regulates a transcriptional program essential for the lissencephaly gene Lis1 and its functional implications. Structure 12, 987 998 (2004). maintenance of hematopoietic stem cells. Nat. Immunol. 12, 129–136 (2011). 23. J. Oberoi et al., Structural basis for the assembly of the SMRT/NCoR core transcrip- – 7. P. E. Love, C. Warzecha, L. Li, Ldb1 complexes: The new master regulators of erythroid tional repression machinery. Nat. Struct. Mol. Biol. 18, 177 184 (2011). gene transcription. Trends Genet. 30,1–9 (2014). 24. H. Wang, Z. Wang, Q. Tang, X. X. Yan, W. Xu, Crystal structure of the LUFS domain of – 8. G. Liu, A. Dean, Enhancer long-range contacts: The multi-adaptor protein LDB1 is the human single-stranded DNA binding protein 2 (SSBP2). Protein Sci. 28,788793 (2019). tie that binds. Biochim. Biophys. Acta. Gene Regul. Mech. 1862, 625–633 (2019). 25. E. Pardon et al., A general protocol for the generation of nanobodies for structural 9. P. Morcillo, C. Rosen, M. K. Baylies, D. Dorsett, Chip, a widely expressed chromosomal

biology. Nat. Protoc. 9, 674–693 (2014). BIOCHEMISTRY protein required for segmentation and activity of a remote wing margin enhancer in 26. A. Plückthun, Ribosome display: A perspective. Methods Mol. Biol. 805,3–28 (2012). Drosophila. Genes Dev. 11, 2729–2740 (1997). 27. T. Lundqvist et al., Crystal structure of scytalone dehydratase—A disease determinant 10. W. Deng et al., Controlling long-range genomic interactions at a native locus by of the rice pathogen, Magnaporthe grisea. Structure 2, 937–944 (1994). targeted tethering of a looping factor. Cell 149, 1233–1244 (2012). 28. E. M. Westbrook, O. E. Piro, P. B. Sigler, The 6-A crystal structure of delta 5-3-keto- 11. I. Krivega, R. K. Dale, A. Dean, Role of LDB1 in the transition from chromatin looping steroid isomerase. Architecture and location of the active center. J. Biol. Chem. 259, to transcription activation. Genes Dev. 28, 1278–1290 (2014). 9096–9103 (1984). 12. L. W. Jurata, G. N. Gill, Functional analysis of the nuclear LIM domain interactor NLI. 29. T. L. Amyes, J. P. Richard, Specificity in transition state binding: The Pauling model Mol. Cell. Biol. 17, 5688–5698 (1997). revisited. Biochemistry 52, 2021–2035 (2013). 13. A. J. Cross, C. M. Jeffries, J. Trewhella, J. M. Matthews, LIM domain binding proteins 1 30. T. L. Bullock, W. D. Clarkson, H. M. Kent, M. Stewart, The 1.6 angstroms resolution – and 2 have different oligomeric states. J. Mol. Biol. 399, 133 144 (2010). crystal structure of nuclear transport factor 2 (NTF2). J. Mol. Biol. 260, 422–431 (1996). 14. D. J. van Meyel, J. B. Thomas, A. D. Agulnick, Ssdp proteins bind to LIM-interacting co- 31. N. Meier et al., Novel binding partners of Ldb1 are required for haematopoietic de- factors and regulate the activity of LIM-homeodomain protein complexes in vivo. velopment. Development 133, 4913–4923 (2006). – Development 130, 1915 1925 (2003). 32. I. A. Wadman et al., The LIM-only protein Lmo2 is a bridging molecule assembling an 15. B. Lee, S. Lee, A. D. Agulnick, J. W. Lee, S. K. Lee, Single-stranded DNA binding erythroid, DNA-binding complex which includes the TAL1, E47, GATA-1 and Ldb1/NLI proteins are required for LIM complexes to induce transcriptionally active chromatin proteins. EMBO J. 16, 3145–3157 (1997). and specify spinal neuronal identities. Development 143, 1721–1731 (2016). 33. T. H. Rabbitts, LMO T-cell translocation oncogenes typify genes activated by chro- 16. R. Bronstein et al., Transcriptional regulation by CHIP/LDB complexes. PLoS Genet. 6, mosomal translocations that alter transcription and developmental processes. Genes e1001063 (2010). Dev. 12, 2651–2657 (1998). 17. M. Fiedler et al., An ancient Pygo-dependent Wnt enhanceosome integrated by Chip/ 34. L. Chen et al., Ssdp proteins interact with the LIM-domain-binding protein Ldb1 to LDB-SSDP. eLife 4, e09073 (2015). regulate development. Proc. Natl. Acad. Sci. U.S.A. 99, 14320–14325 (2002). 18. J. Li et al., Requirement for ssbp2 in hematopoietic stem cell maintenance and stress 35. N. Nishioka et al., Ssdp1 regulates head morphogenesis of mouse embryos by acti- response. J. Immunol. 193, 4654–4662 (2014). vating the Lim1-Ldb1 complex. Development 132, 2535–2546 (2005).

Renko et al. PNAS | October 15, 2019 | vol. 116 | no. 42 | 20983 Downloaded by guest on September 24, 2021